[DERCBOT-919] Add condenseQuestion LLM and Prompt

Author: assouktim

bot/admin/server/src/main/kotlin/BotAdminService.kt

@@ -1156,7 +1156,9 @@ object BotAdminService {
         // delete the RAG configuration
         ragConfigurationDAO.findByNamespaceAndBotId(app.namespace, app.name)?.let { config ->
             ragConfigurationDAO.delete(config._id)
-            config.llmSetting.apiKey?.let { SecurityUtils.deleteSecret(it) }
+            config.questionCondensingLlmSetting?.apiKey?.let { SecurityUtils.deleteSecret(it) }
+            config.questionAnsweringLlmSetting?.apiKey?.let { SecurityUtils.deleteSecret(it) }
+            config.llmSetting?.apiKey?.let { SecurityUtils.deleteSecret(it) }
             config.emSetting.apiKey?.let { SecurityUtils.deleteSecret(it) }
         }
 

bot/admin/server/src/main/kotlin/model/BotRAGConfigurationDTO.kt

@@ -18,6 +18,7 @@ package ai.tock.bot.admin.model
 
 import ai.tock.bot.admin.bot.rag.BotRAGConfiguration
 import ai.tock.bot.admin.service.VectorStoreService
+import ai.tock.genai.orchestratorclient.requests.PromptTemplate
 import ai.tock.genai.orchestratorcore.mappers.EMSettingMapper
 import ai.tock.genai.orchestratorcore.mappers.LLMSettingMapper
 import ai.tock.genai.orchestratorcore.models.Constants
@@ -34,26 +35,39 @@ data class BotRAGConfigurationDTO(
     val namespace: String,
     val botId: String,
     val enabled: Boolean = false,
-    val llmSetting: LLMSettingDTO,
+    val questionCondensingLlmSetting: LLMSettingDTO? = null,
+    val questionCondensingPrompt: PromptTemplate? = null,
+    val questionAnsweringLlmSetting: LLMSettingDTO,
+    val questionAnsweringPrompt: PromptTemplate,
     val emSetting: EMSettingDTO,
     val indexSessionId: String? = null,
     val indexName: String? = null,
     val noAnswerSentence: String,
     val noAnswerStoryId: String? = null,
     val documentsRequired: Boolean = true,
+    val debugEnabled: Boolean,
+    val maxDocumentsRetrieved: Int,
+    val maxMessagesFromHistory: Int,
 ) {
     constructor(configuration: BotRAGConfiguration) : this(
         id = configuration._id.toString(),
         namespace = configuration.namespace,
         botId = configuration.botId,
         enabled = configuration.enabled,
-        llmSetting = configuration.llmSetting.toDTO(),
+        questionCondensingLlmSetting = configuration.questionCondensingLlmSetting?.toDTO(),
+        questionCondensingPrompt = configuration.questionCondensingPrompt,
+        questionAnsweringLlmSetting = configuration.getQuestionAnsweringLLMSetting().toDTO(),
+        questionAnsweringPrompt = configuration.questionAnsweringPrompt
+            ?: configuration.initQuestionAnsweringPrompt(),
         emSetting = configuration.emSetting.toDTO(),
         indexSessionId = configuration.indexSessionId,
         indexName = configuration.generateIndexName(),
         noAnswerSentence = configuration.noAnswerSentence,
         noAnswerStoryId = configuration.noAnswerStoryId,
         documentsRequired = configuration.documentsRequired,
+        debugEnabled = configuration.debugEnabled,
+        maxDocumentsRetrieved = configuration.maxDocumentsRetrieved,
+        maxMessagesFromHistory = configuration.maxMessagesFromHistory,
     )
 
     fun toBotRAGConfiguration(): BotRAGConfiguration =
@@ -62,12 +76,20 @@ data class BotRAGConfigurationDTO(
             namespace = namespace,
             botId = botId,
             enabled = enabled,
-            llmSetting = LLMSettingMapper.toEntity(
+            questionCondensingLlmSetting = LLMSettingMapper.toEntity(
+                namespace = namespace,
+                botId = botId,
+                feature = Constants.GEN_AI_RAG_QUESTION_CONDENSING,
+                dto = questionCondensingLlmSetting!!
+            ),
+            questionCondensingPrompt = questionCondensingPrompt,
+            questionAnsweringLlmSetting = LLMSettingMapper.toEntity(
                 namespace = namespace,
                 botId = botId,
                 feature = Constants.GEN_AI_RAG_QUESTION_ANSWERING,
-                dto = llmSetting
+                dto = questionAnsweringLlmSetting
             ),
+            questionAnsweringPrompt = questionAnsweringPrompt,
             emSetting = EMSettingMapper.toEntity(
                 namespace = namespace,
                 botId = botId,
@@ -78,6 +100,9 @@ data class BotRAGConfigurationDTO(
             noAnswerSentence = noAnswerSentence,
             noAnswerStoryId = noAnswerStoryId,
             documentsRequired = documentsRequired,
+            debugEnabled = debugEnabled,
+            maxDocumentsRetrieved = maxDocumentsRetrieved,
+            maxMessagesFromHistory = maxMessagesFromHistory,
         )
 }
 
@@ -87,6 +112,7 @@ private fun BotRAGConfiguration.generateIndexName(): String? {
             namespace,
             botId,
             it,
+            maxDocumentsRetrieved,
             VectorStoreService.getVectorStoreConfiguration(namespace, botId, enabled = true)
                 ?.setting
         ).second

bot/admin/server/src/main/kotlin/model/BotSentenceGenerationConfigurationDTO.kt

@@ -17,6 +17,7 @@
 package ai.tock.bot.admin.model
 
 import ai.tock.bot.admin.bot.sentencegeneration.BotSentenceGenerationConfiguration
+import ai.tock.genai.orchestratorclient.requests.PromptTemplate
 import ai.tock.genai.orchestratorcore.mappers.LLMSettingMapper
 import ai.tock.genai.orchestratorcore.models.Constants
 import ai.tock.genai.orchestratorcore.models.llm.LLMSettingDTO
@@ -32,6 +33,7 @@ data class BotSentenceGenerationConfigurationDTO(
     val enabled: Boolean = false,
     val nbSentences: Int,
     val llmSetting: LLMSettingDTO,
+    val prompt: PromptTemplate,
 ) {
     constructor(configuration: BotSentenceGenerationConfiguration) : this(
         id = configuration._id.toString(),
@@ -40,6 +42,7 @@ data class BotSentenceGenerationConfigurationDTO(
         enabled = configuration.enabled,
         nbSentences = configuration.nbSentences,
         llmSetting = configuration.llmSetting.toDTO(),
+        prompt = configuration.prompt ?: configuration.initPrompt()
     )
 
     fun toSentenceGenerationConfiguration(): BotSentenceGenerationConfiguration =
@@ -54,7 +57,8 @@ data class BotSentenceGenerationConfigurationDTO(
                 botId = botId,
                 feature = Constants.GEN_AI_COMPLETION_SENTENCE_GENERATION,
                 dto = llmSetting
-            )
+            ),
+            prompt = prompt
         )
 }
 

bot/admin/server/src/main/kotlin/model/GenerateSentencesRequest.kt

@@ -16,9 +16,11 @@
 
 package ai.tock.bot.admin.model
 
+import ai.tock.genai.orchestratorclient.requests.PromptTemplate
 import java.util.Locale
 
 data class SentenceGenerationRequest(
+    val prompt: PromptTemplate,
     val llmTemperature: String,
     val sentences: List<String>,
     val locale: Locale,

bot/admin/server/src/main/kotlin/service/CompletionService.kt

@@ -75,18 +75,11 @@ object CompletionService {
             )
         )
 
-        // Create a Jinja2 prompt template
-        val prompt = PromptTemplate(
-            formatter = Formatter.JINJA2.id,
-            template = llmSetting.prompt,
-            inputs = inputs
-        )
-
         // call the completion service to generate sentences
         return completionService
             .generateSentences(
                 SentenceGenerationQuery(
-                    llmSetting, prompt,
+                    llmSetting, request.prompt.copy(inputs = inputs),
                     ObservabilityService.getObservabilityConfiguration(namespace, botId, enabled = true)?.setting
                 )
             )

bot/admin/server/src/main/kotlin/service/RAGService.kt

@@ -64,8 +64,11 @@ object RAGService {
         logger.info { "Deleting the RAG Configuration [namespace: $namespace, botId: $botId]" }
         ragConfigurationDAO.delete(ragConfig._id)
 
-        logger.info { "Deleting the LLM secret ..." }
-        ragConfig.llmSetting.apiKey?.let { SecurityUtils.deleteSecret(it) }
+        logger.info { "Deleting the question condensing LLM secret ..." }
+        ragConfig.questionCondensingLlmSetting?.apiKey?.let { SecurityUtils.deleteSecret(it) }
+        logger.info { "Deleting the question answering LLM secret ..." }
+        ragConfig.questionAnsweringLlmSetting?.apiKey?.let { SecurityUtils.deleteSecret(it) }
+        ragConfig.llmSetting?.apiKey?.let { SecurityUtils.deleteSecret(it) }
         logger.info { "Deleting the Embedding secret ..." }
         ragConfig.emSetting.apiKey?.let { SecurityUtils.deleteSecret(it) }
     }

bot/admin/server/src/main/kotlin/service/RAGValidationService.kt

@@ -37,15 +37,24 @@ object RAGValidationService {
     private val vectorStoreProviderService: VectorStoreProviderService get() = injector.provide()
 
     fun validate(ragConfig: BotRAGConfiguration): Set<ErrorMessage> {
+        val observabilitySetting = ObservabilityService.getObservabilityConfiguration(
+            ragConfig.namespace, ragConfig.botId, enabled = true
+        )?.setting
+
         return mutableSetOf<ErrorMessage>().apply {
-            val llmErrors = llmProviderService.checkSetting(
+            val questionCondensingLlmErrors = llmProviderService.checkSetting(
+                LLMProviderSettingStatusQuery(
+                    ragConfig.questionCondensingLlmSetting!!,
+                    observabilitySetting
+                )
+            ).getErrors("LLM setting check failed (for question condensing)")
+
+            val questionAnsweringLlmErrors = llmProviderService.checkSetting(
                 LLMProviderSettingStatusQuery(
-                    ragConfig.llmSetting,
-                    ObservabilityService.getObservabilityConfiguration(
-                        ragConfig.namespace, ragConfig.botId, enabled = true
-                    )?.setting
+                    ragConfig.questionAnsweringLlmSetting!!,
+                    observabilitySetting
                 )
-            ).getErrors("LLM setting check failed")
+            ).getErrors("LLM setting check failed (for question answering)")
 
             val embeddingErrors = emProviderService.checkSetting(
                 EMProviderSettingStatusQuery(ragConfig.emSetting)
@@ -59,7 +68,11 @@ object RAGValidationService {
                 )?.setting
 
                 val (_, indexName) = VectorStoreUtils.getVectorStoreElements(
-                    ragConfig.namespace, ragConfig.botId, ragConfig.indexSessionId!!, vectorStoreSetting
+                    ragConfig.namespace,
+                    ragConfig.botId,
+                    ragConfig.indexSessionId!!,
+                    ragConfig.maxDocumentsRetrieved,
+                    vectorStoreSetting
                 )
 
                 vectorStoreProviderService.checkSetting(
@@ -71,7 +84,7 @@ object RAGValidationService {
                 ).getErrors("Vector store setting check failed")
             } ?: emptySet()
 
-            addAll(llmErrors + embeddingErrors + indexSessionIdErrors + vectorStoreErrors)
+            addAll(questionCondensingLlmErrors + questionAnsweringLlmErrors + embeddingErrors + indexSessionIdErrors + vectorStoreErrors)
         }
     }
 

bot/admin/server/src/test/kotlin/service/RAGServiceTest.kt

@@ -30,6 +30,7 @@ import ai.tock.bot.test.TFunction
 import ai.tock.bot.test.TRunnable
 import ai.tock.bot.test.TSupplier
 import ai.tock.bot.test.TestCase
+import ai.tock.genai.orchestratorclient.requests.PromptTemplate
 import ai.tock.genai.orchestratorclient.responses.ProviderSettingStatusResponse
 import ai.tock.genai.orchestratorclient.services.EMProviderService
 import ai.tock.genai.orchestratorclient.services.LLMProviderService
@@ -65,25 +66,36 @@ class RAGServiceTest : AbstractTest() {
         const val INDEX_SESSION_ID = "1010101"
 
         private val DEFAULT_RAG_CONFIG = BotRAGConfigurationDTO(
-            id = "ragId",
+             id = "ragId",
             namespace = NAMESPACE,
             botId = BOT_ID,
             enabled = false,
-            llmSetting = OpenAILLMSettingDTO(
+            questionCondensingLlmSetting = OpenAILLMSettingDTO(
                 apiKey = "apikey",
                 model = MODEL,
-                prompt = PROMPT,
                 temperature = TEMPERATURE,
                 baseUrl = "https://api.openai.com/v1"
             ),
+            questionCondensingPrompt = PromptTemplate(template = PROMPT),
+            questionAnsweringLlmSetting = OpenAILLMSettingDTO(
+                apiKey = "apikey",
+                model = MODEL,
+                temperature = TEMPERATURE,
+                baseUrl = "https://api.openai.com/v1"
+            ),
+            questionAnsweringPrompt = PromptTemplate(template = PROMPT),
             emSetting = AzureOpenAIEMSettingDTO(
                 apiKey = "apiKey",
                 apiVersion = "apiVersion",
                 deploymentName = "deployment",
                 model = "model",
                 apiBase = "url"
             ),
-            noAnswerSentence = "No answer sentence"
+            noAnswerSentence = "No answer sentence",
+            documentsRequired = true,
+            debugEnabled = false,
+            maxDocumentsRetrieved = 2,
+            maxMessagesFromHistory = 2,
         )
 
         private val DEFAULT_BOT_CONFIG = aApplication.copy(namespace = NAMESPACE, botId = BOT_ID)
@@ -186,9 +198,9 @@ class RAGServiceTest : AbstractTest() {
             Assertions.assertEquals(BOT_ID, captured.botId)
             Assertions.assertEquals(true, captured.enabled)
             Assertions.assertEquals(NAMESPACE, captured.namespace)
-            Assertions.assertEquals(PROVIDER, captured.llmSetting.provider.name)
-            Assertions.assertEquals(TEMPERATURE, captured.llmSetting.temperature)
-            Assertions.assertEquals(PROMPT, captured.llmSetting.prompt)
+            Assertions.assertEquals(PROVIDER, captured.questionAnsweringLlmSetting!!.provider.name)
+            Assertions.assertEquals(TEMPERATURE, captured.questionAnsweringLlmSetting!!.temperature)
+            Assertions.assertEquals(PROMPT, captured.questionAnsweringPrompt!!.template)
             Assertions.assertEquals(null, captured.noAnswerStoryId)
         }
 

bot/admin/server/src/test/kotlin/service/RAGValidationServiceTest.kt

@@ -19,6 +19,7 @@ package ai.tock.bot.admin.service
 import ai.tock.bot.admin.bot.observability.BotObservabilityConfigurationDAO
 import ai.tock.bot.admin.bot.vectorstore.BotVectorStoreConfigurationDAO
 import ai.tock.bot.admin.model.BotRAGConfigurationDTO
+import ai.tock.genai.orchestratorclient.requests.PromptTemplate
 import ai.tock.genai.orchestratorclient.responses.ErrorInfo
 import ai.tock.genai.orchestratorclient.responses.ErrorResponse
 import ai.tock.genai.orchestratorclient.responses.ProviderSettingStatusResponse
@@ -63,7 +64,7 @@ class RAGValidationServiceTest {
     }
 
     private val openAILLMSetting = OpenAILLMSetting(
-        apiKey = "123-abc", model = "unavailable-model", temperature = "0.4", prompt = "How to bike in the rain",
+        apiKey = "123-abc", model = "unavailable-model", temperature = "0.4",
         baseUrl = "https://api.openai.com/v1",
     )
 
@@ -78,9 +79,16 @@ class RAGValidationServiceTest {
     private val ragConfiguration = BotRAGConfigurationDTO(
         namespace = "namespace",
         botId = "botId",
-        llmSetting = openAILLMSetting,
+        questionCondensingLlmSetting = openAILLMSetting,
+        questionCondensingPrompt = PromptTemplate(template = "test"),
+        questionAnsweringLlmSetting = openAILLMSetting,
+        questionAnsweringPrompt = PromptTemplate(template = "How to bike in the rain"),
         emSetting = azureOpenAIEMSetting,
         noAnswerSentence = " No answer sentence",
+        documentsRequired = true,
+        debugEnabled = false,
+        maxDocumentsRetrieved = 2,
+        maxMessagesFromHistory = 2,
     )
 
     @Test
@@ -163,7 +171,7 @@ class RAGValidationServiceTest {
     fun `validation of the RAG configuration when the Orchestrator returns 2 errors for LLM and 1 for Embedding model, the RAG function has not been activated`() {
 
         // GIVEN
-        // - 3 errors returned by Generative AI Orchestrator for LLM (2) and EM (1)
+        // - 3 errors returned by Generative AI Orchestrator for LLM (4 = 2 for condensing + 2 for answering) and EM (1)
         // - RAG is not enabled
         every {
             llmProviderService.checkSetting(any())
@@ -187,11 +195,13 @@ class RAGValidationServiceTest {
         )
 
         // THEN :
-        // Check that 3 errors have been found
-        assertEquals(2, errors.size)
+        // Check that 3 groups of errors have been found
+        assertEquals(3, errors.size)
         assertEquals("10", (((errors.elementAt(0).params) as List<*>)[0] as ErrorResponse).code)
         assertEquals("20", (((errors.elementAt(0).params) as List<*>)[1] as ErrorResponse).code)
-        assertEquals("30", (((errors.elementAt(1).params) as List<*>)[0] as ErrorResponse).code)
+        assertEquals("10", (((errors.elementAt(1).params) as List<*>)[0] as ErrorResponse).code)
+        assertEquals("20", (((errors.elementAt(1).params) as List<*>)[1] as ErrorResponse).code)
+        assertEquals("30", (((errors.elementAt(2).params) as List<*>)[0] as ErrorResponse).code)
     }
 
     private fun createFakeErrorResponse(code: String) = ErrorResponse(